1 /* 2 * inet fragments management 3 * 4 * This program is free software; you can redistribute it and/or 5 * modify it under the terms of the GNU General Public License 6 * as published by the Free Software Foundation; either version 7 * 2 of the License, or (at your option) any later version. 8 * 9 * Authors: Pavel Emelyanov <xemul@openvz.org> 10 * Started as consolidation of ipv4/ip_fragment.c, 11 * ipv6/reassembly. and ipv6 nf conntrack reassembly 12 */ 13 14 #include <linux/list.h> 15 #include <linux/spinlock.h> 16 #include <linux/module.h> 17 #include <linux/timer.h> 18 #include <linux/mm.h> 19 #include <linux/random.h> 20 #include <linux/skbuff.h> 21 #include <linux/rtnetlink.h> 22 #include <linux/slab.h> 23 24 #include <net/sock.h> 25 #include <net/inet_frag.h> 26 #include <net/inet_ecn.h> 27 28 #define INETFRAGS_EVICT_BUCKETS 128 29 #define INETFRAGS_EVICT_MAX 512 30 31 /* don't rebuild inetfrag table with new secret more often than this */ 32 #define INETFRAGS_MIN_REBUILD_INTERVAL (5 * HZ) 33 34 /* Given the OR values of all fragments, apply RFC 3168 5.3 requirements 35 * Value : 0xff if frame should be dropped. 36 * 0 or INET_ECN_CE value, to be ORed in to final iph->tos field 37 */ 38 const u8 ip_frag_ecn_table[16] = { 39 /* at least one fragment had CE, and others ECT_0 or ECT_1 */ 40 [IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0] = INET_ECN_CE, 41 [IPFRAG_ECN_CE | IPFRAG_ECN_ECT_1] = INET_ECN_CE, 42 [IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0 | IPFRAG_ECN_ECT_1] = INET_ECN_CE, 43 44 /* invalid combinations : drop frame */ 45 [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE] = 0xff, 46 [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_ECT_0] = 0xff, 47 [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_ECT_1] = 0xff, 48 [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_ECT_0 | IPFRAG_ECN_ECT_1] = 0xff, 49 [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0] = 0xff, 50 [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE | IPFRAG_ECN_ECT_1] = 0xff, 51 [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0 | IPFRAG_ECN_ECT_1] = 0xff, 52 }; 53 EXPORT_SYMBOL(ip_frag_ecn_table); 54 55 static unsigned int 56 inet_frag_hashfn(const struct inet_frags *f, const struct inet_frag_queue *q) 57 { 58 return f->hashfn(q) & (INETFRAGS_HASHSZ - 1); 59 } 60 61 static bool inet_frag_may_rebuild(struct inet_frags *f) 62 { 63 return time_after(jiffies, 64 f->last_rebuild_jiffies + INETFRAGS_MIN_REBUILD_INTERVAL); 65 } 66 67 static void inet_frag_secret_rebuild(struct inet_frags *f) 68 { 69 int i; 70 71 write_seqlock_bh(&f->rnd_seqlock); 72 73 if (!inet_frag_may_rebuild(f)) 74 goto out; 75 76 get_random_bytes(&f->rnd, sizeof(u32)); 77 78 for (i = 0; i < INETFRAGS_HASHSZ; i++) { 79 struct inet_frag_bucket *hb; 80 struct inet_frag_queue *q; 81 struct hlist_node *n; 82 83 hb = &f->hash[i]; 84 spin_lock(&hb->chain_lock); 85 86 hlist_for_each_entry_safe(q, n, &hb->chain, list) { 87 unsigned int hval = inet_frag_hashfn(f, q); 88 89 if (hval != i) { 90 struct inet_frag_bucket *hb_dest; 91 92 hlist_del(&q->list); 93 94 /* Relink to new hash chain. */ 95 hb_dest = &f->hash[hval]; 96 97 /* This is the only place where we take 98 * another chain_lock while already holding 99 * one. As this will not run concurrently, 100 * we cannot deadlock on hb_dest lock below, if its 101 * already locked it will be released soon since 102 * other caller cannot be waiting for hb lock 103 * that we've taken above. 104 */ 105 spin_lock_nested(&hb_dest->chain_lock, 106 SINGLE_DEPTH_NESTING); 107 hlist_add_head(&q->list, &hb_dest->chain); 108 spin_unlock(&hb_dest->chain_lock); 109 } 110 } 111 spin_unlock(&hb->chain_lock); 112 } 113 114 f->rebuild = false; 115 f->last_rebuild_jiffies = jiffies; 116 out: 117 write_sequnlock_bh(&f->rnd_seqlock); 118 } 119 120 static bool inet_fragq_should_evict(const struct inet_frag_queue *q) 121 { 122 if (!hlist_unhashed(&q->list_evictor)) 123 return false; 124 125 return q->net->low_thresh == 0 || 126 frag_mem_limit(q->net) >= q->net->low_thresh; 127 } 128 129 static unsigned int 130 inet_evict_bucket(struct inet_frags *f, struct inet_frag_bucket *hb) 131 { 132 struct inet_frag_queue *fq; 133 struct hlist_node *n; 134 unsigned int evicted = 0; 135 HLIST_HEAD(expired); 136 137 spin_lock(&hb->chain_lock); 138 139 hlist_for_each_entry_safe(fq, n, &hb->chain, list) { 140 if (!inet_fragq_should_evict(fq)) 141 continue; 142 143 if (!del_timer(&fq->timer)) 144 continue; 145 146 hlist_add_head(&fq->list_evictor, &expired); 147 ++evicted; 148 } 149 150 spin_unlock(&hb->chain_lock); 151 152 hlist_for_each_entry_safe(fq, n, &expired, list_evictor) 153 f->frag_expire(&fq->timer); 154 155 return evicted; 156 } 157 158 static void inet_frag_worker(struct work_struct *work) 159 { 160 unsigned int budget = INETFRAGS_EVICT_BUCKETS; 161 unsigned int i, evicted = 0; 162 struct inet_frags *f; 163 164 f = container_of(work, struct inet_frags, frags_work); 165 166 BUILD_BUG_ON(INETFRAGS_EVICT_BUCKETS >= INETFRAGS_HASHSZ); 167 168 local_bh_disable(); 169 170 for (i = READ_ONCE(f->next_bucket); budget; --budget) { 171 evicted += inet_evict_bucket(f, &f->hash[i]); 172 i = (i + 1) & (INETFRAGS_HASHSZ - 1); 173 if (evicted > INETFRAGS_EVICT_MAX) 174 break; 175 } 176 177 f->next_bucket = i; 178 179 local_bh_enable(); 180 181 if (f->rebuild && inet_frag_may_rebuild(f)) 182 inet_frag_secret_rebuild(f); 183 } 184 185 static void inet_frag_schedule_worker(struct inet_frags *f) 186 { 187 if (unlikely(!work_pending(&f->frags_work))) 188 schedule_work(&f->frags_work); 189 } 190 191 int inet_frags_init(struct inet_frags *f) 192 { 193 int i; 194 195 INIT_WORK(&f->frags_work, inet_frag_worker); 196 197 for (i = 0; i < INETFRAGS_HASHSZ; i++) { 198 struct inet_frag_bucket *hb = &f->hash[i]; 199 200 spin_lock_init(&hb->chain_lock); 201 INIT_HLIST_HEAD(&hb->chain); 202 } 203 204 seqlock_init(&f->rnd_seqlock); 205 f->last_rebuild_jiffies = 0; 206 f->frags_cachep = kmem_cache_create(f->frags_cache_name, f->qsize, 0, 0, 207 NULL); 208 if (!f->frags_cachep) 209 return -ENOMEM; 210 211 return 0; 212 } 213 EXPORT_SYMBOL(inet_frags_init); 214 215 void inet_frags_fini(struct inet_frags *f) 216 { 217 cancel_work_sync(&f->frags_work); 218 kmem_cache_destroy(f->frags_cachep); 219 } 220 EXPORT_SYMBOL(inet_frags_fini); 221 222 void inet_frags_exit_net(struct netns_frags *nf, struct inet_frags *f) 223 { 224 unsigned int seq; 225 int i; 226 227 nf->low_thresh = 0; 228 229 evict_again: 230 local_bh_disable(); 231 seq = read_seqbegin(&f->rnd_seqlock); 232 233 for (i = 0; i < INETFRAGS_HASHSZ ; i++) 234 inet_evict_bucket(f, &f->hash[i]); 235 236 local_bh_enable(); 237 cond_resched(); 238 239 if (read_seqretry(&f->rnd_seqlock, seq) || 240 sum_frag_mem_limit(nf)) 241 goto evict_again; 242 } 243 EXPORT_SYMBOL(inet_frags_exit_net); 244 245 static struct inet_frag_bucket * 246 get_frag_bucket_locked(struct inet_frag_queue *fq, struct inet_frags *f) 247 __acquires(hb->chain_lock) 248 { 249 struct inet_frag_bucket *hb; 250 unsigned int seq, hash; 251 252 restart: 253 seq = read_seqbegin(&f->rnd_seqlock); 254 255 hash = inet_frag_hashfn(f, fq); 256 hb = &f->hash[hash]; 257 258 spin_lock(&hb->chain_lock); 259 if (read_seqretry(&f->rnd_seqlock, seq)) { 260 spin_unlock(&hb->chain_lock); 261 goto restart; 262 } 263 264 return hb; 265 } 266 267 static inline void fq_unlink(struct inet_frag_queue *fq, struct inet_frags *f) 268 { 269 struct inet_frag_bucket *hb; 270 271 hb = get_frag_bucket_locked(fq, f); 272 hlist_del(&fq->list); 273 fq->flags |= INET_FRAG_COMPLETE; 274 spin_unlock(&hb->chain_lock); 275 } 276 277 void inet_frag_kill(struct inet_frag_queue *fq, struct inet_frags *f) 278 { 279 if (del_timer(&fq->timer)) 280 refcount_dec(&fq->refcnt); 281 282 if (!(fq->flags & INET_FRAG_COMPLETE)) { 283 fq_unlink(fq, f); 284 refcount_dec(&fq->refcnt); 285 } 286 } 287 EXPORT_SYMBOL(inet_frag_kill); 288 289 void inet_frag_destroy(struct inet_frag_queue *q, struct inet_frags *f) 290 { 291 struct sk_buff *fp; 292 struct netns_frags *nf; 293 unsigned int sum, sum_truesize = 0; 294 295 WARN_ON(!(q->flags & INET_FRAG_COMPLETE)); 296 WARN_ON(del_timer(&q->timer) != 0); 297 298 /* Release all fragment data. */ 299 fp = q->fragments; 300 nf = q->net; 301 while (fp) { 302 struct sk_buff *xp = fp->next; 303 304 sum_truesize += fp->truesize; 305 kfree_skb(fp); 306 fp = xp; 307 } 308 sum = sum_truesize + f->qsize; 309 310 if (f->destructor) 311 f->destructor(q); 312 kmem_cache_free(f->frags_cachep, q); 313 314 sub_frag_mem_limit(nf, sum); 315 } 316 EXPORT_SYMBOL(inet_frag_destroy); 317 318 static struct inet_frag_queue *inet_frag_intern(struct netns_frags *nf, 319 struct inet_frag_queue *qp_in, 320 struct inet_frags *f, 321 void *arg) 322 { 323 struct inet_frag_bucket *hb = get_frag_bucket_locked(qp_in, f); 324 struct inet_frag_queue *qp; 325 326 #ifdef CONFIG_SMP 327 /* With SMP race we have to recheck hash table, because 328 * such entry could have been created on other cpu before 329 * we acquired hash bucket lock. 330 */ 331 hlist_for_each_entry(qp, &hb->chain, list) { 332 if (qp->net == nf && f->match(qp, arg)) { 333 refcount_inc(&qp->refcnt); 334 spin_unlock(&hb->chain_lock); 335 qp_in->flags |= INET_FRAG_COMPLETE; 336 inet_frag_put(qp_in, f); 337 return qp; 338 } 339 } 340 #endif 341 qp = qp_in; 342 if (!mod_timer(&qp->timer, jiffies + nf->timeout)) 343 refcount_inc(&qp->refcnt); 344 345 refcount_inc(&qp->refcnt); 346 hlist_add_head(&qp->list, &hb->chain); 347 348 spin_unlock(&hb->chain_lock); 349 350 return qp; 351 } 352 353 static struct inet_frag_queue *inet_frag_alloc(struct netns_frags *nf, 354 struct inet_frags *f, 355 void *arg) 356 { 357 struct inet_frag_queue *q; 358 359 if (!nf->high_thresh || frag_mem_limit(nf) > nf->high_thresh) { 360 inet_frag_schedule_worker(f); 361 return NULL; 362 } 363 364 q = kmem_cache_zalloc(f->frags_cachep, GFP_ATOMIC); 365 if (!q) 366 return NULL; 367 368 q->net = nf; 369 f->constructor(q, arg); 370 add_frag_mem_limit(nf, f->qsize); 371 372 timer_setup(&q->timer, f->frag_expire, 0); 373 spin_lock_init(&q->lock); 374 refcount_set(&q->refcnt, 1); 375 376 return q; 377 } 378 379 static struct inet_frag_queue *inet_frag_create(struct netns_frags *nf, 380 struct inet_frags *f, 381 void *arg) 382 { 383 struct inet_frag_queue *q; 384 385 q = inet_frag_alloc(nf, f, arg); 386 if (!q) 387 return NULL; 388 389 return inet_frag_intern(nf, q, f, arg); 390 } 391 392 struct inet_frag_queue *inet_frag_find(struct netns_frags *nf, 393 struct inet_frags *f, void *key, 394 unsigned int hash) 395 { 396 struct inet_frag_bucket *hb; 397 struct inet_frag_queue *q; 398 int depth = 0; 399 400 if (frag_mem_limit(nf) > nf->low_thresh) 401 inet_frag_schedule_worker(f); 402 403 hash &= (INETFRAGS_HASHSZ - 1); 404 hb = &f->hash[hash]; 405 406 spin_lock(&hb->chain_lock); 407 hlist_for_each_entry(q, &hb->chain, list) { 408 if (q->net == nf && f->match(q, key)) { 409 refcount_inc(&q->refcnt); 410 spin_unlock(&hb->chain_lock); 411 return q; 412 } 413 depth++; 414 } 415 spin_unlock(&hb->chain_lock); 416 417 if (depth <= INETFRAGS_MAXDEPTH) 418 return inet_frag_create(nf, f, key); 419 420 if (inet_frag_may_rebuild(f)) { 421 if (!f->rebuild) 422 f->rebuild = true; 423 inet_frag_schedule_worker(f); 424 } 425 426 return ERR_PTR(-ENOBUFS); 427 } 428 EXPORT_SYMBOL(inet_frag_find); 429 430 void inet_frag_maybe_warn_overflow(struct inet_frag_queue *q, 431 const char *prefix) 432 { 433 static const char msg[] = "inet_frag_find: Fragment hash bucket" 434 " list length grew over limit " __stringify(INETFRAGS_MAXDEPTH) 435 ". Dropping fragment.\n"; 436 437 if (PTR_ERR(q) == -ENOBUFS) 438 net_dbg_ratelimited("%s%s", prefix, msg); 439 } 440 EXPORT_SYMBOL(inet_frag_maybe_warn_overflow); 441